How Rising AI Token Costs Are Reshaping Marketing Agency Economics

Marketing agency S4 Capital doubled its operating profit in the first half of 2024 to £35.2 million through strict expense discipline, but faced a new challenge: AI token consumption in the Monks division is growing exponentially, and Goldman Sachs forecasts indicate a 24-fold increase in token consumption volume between 2026 and 2030. This is forcing agencies to rethink their generative artificial intelligence workflows and seek balance between technology investments and cost control.

Why AI costs are becoming a critical expense line

Monks founder and chief AI officer Wesley ter Haar states that token usage is "exploding." This is predictable for an agency with a client base of technology brands and years of investment in generative tools, but it creates financial pressure. S4 Capital chairman Sir Martin Sorrell acknowledged that the company "probably wasn't disciplined enough" in controlling these expenses.

The expense structure is changing dramatically. WPP CFO Joanne Wilson notes that while agency service costs were previously formed almost entirely from personnel costs, a second component has now emerged—technology expenses. This requires a rethinking of commercial models and pricing approaches.

How agencies control token expenses

Independent agency PMG from Texas set a strict limit: $50 per employee per day for token usage as of May 2024. This is a proactive step—Forrester research shows that 60% of agencies focus on paying for third-party tools via SaaS model or per user seat, while only 35% pay attention to cloud and computing expenses.

Monks chose a more flexible approach. Ter Haar proposes a "one size fits one" model: limits are set for teams using AI agents for routine tasks, but removed for expert areas like programming, where token consumption has grown most significantly with the spread of AI agents. Monks' flagship platform Monks.flow automatically switches between models depending on the task, factoring in token efficiency, while some employees' access is limited to pre-configured workflows with predictable expenses.

"If you let the whole team maximize token usage, you'll end up with a group of people burning through your budget trying to figure out what the weather will be next week"

Risks of investors subsidizing AI tools

Forrester analyst Jay Pattisall warns that current AI tool availability is artificially cheap. "The reason all this is possible now is that stock markets and private investment are subsidizing AI costs," he explains. Once investors realize the business model limitations and providers shift from SaaS subscriptions to token or GPU-based pricing, the real cost of AI will rise significantly—possibly to the point where advertising automation becomes more expensive than hiring specialists.

This scenario could be devastating for the marketing agency industry, which has cut approximately 8% of staff in 2025 and the first half of 2026. S4 Capital has reduced headcount from 8,200 people four years ago to 6,150 now. Forrester research highlights a monetization problem: only 9% of agencies generate direct revenue from generative AI, while 61% view it as part of basic business operating costs.

Alternative commercial models to offset expenses

Monks includes token costs in overall project pricing, and for Monks.flow clients passes them through directly without markup. The division is actively developing a subscription model: clients receive a fixed number of outputs and regular updates for a constant fee. According to ter Haar's forecast, subscriptions will account for 25% of revenue by year-end, and almost all new contracts are being signed either on subscription or with outcome-based pricing.

Outcome-based pricing models theoretically allow revenue to scale with token growth. WPP is also gradually rolling out outcome-based pricing, adapting cost calculations to the new expense structure. However, ter Haar notes a barrier: while marketing directors intuitively understand the value of outcome-based agreements, corporate procurement departments involved in tenders and RFPs prefer comparing fixed rates. "Clients don't yet know how to buy services on variable commercial models," he states.

What this means for the Russian market

Russian agencies and brands will face similar challenges with a time lag. Rising costs of computing power and tokens will affect all market participants working with AI tools for content creation, media planning, and analytics. At the same time, the local market has its own specifics: the dominance of local and adapted solutions rather than Western platforms, which may alter pricing dynamics.

For brands, it is critical now to clarify with contractor agencies how AI expenses are factored into the service cost structure. Transparency in this expense line will help avoid unexpected budget increases in the medium term and assess the actual effectiveness of generative technology use compared to traditional content production methods.

Marketer's checklist: how to control AI costs when working with an agency

• Request from your agency a breakdown of costs: what share of the budget goes to AI tokens, cloud computing, and tool licenses. Demand a separate line item in the estimate.
• Clarify what specific tasks are being solved with AI and evaluate whether automation is truly justified in each case. Routine text variation generation may not require expensive models.
• Discuss the possibility of switching to a hybrid payment model: fixed portion for strategy and basic activities, variable portion for AI generation with transparent token accounting.
• Set KPIs for AI usage efficiency: how many iterations are needed to approve materials, what share of AI-generated content is used without rework, how has production time changed.
• Include in your contract a condition for quarterly review of the AI expense structure with the ability to adjust the approach if an agreed threshold is exceeded.
• Ask the agency to demonstrate how they optimize model selection for different task types—using lighter models for simple operations reduces token consumption without quality loss.

Frequently asked questions

How much does AI token usage cost agencies

Cost depends on the model and provider, but agency PMG set a limit of $50 per employee per day, which indirectly indicates the scale of expenses. Goldman Sachs forecasts a 24-fold increase in token consumption between 2026 and 2030, making this an increasingly critical expense item for agency economics. Currently, many providers subsidize usage through investment, but a shift to full token-based pricing could significantly increase costs.

How do agencies pass AI costs to clients

There are three main models: including token costs in the overall project price (used by most), passing expenses directly without markup (Monks.flow), and subscription models with fixed fees for a certain number of outputs. Monks plans to derive 25% of revenue from subscriptions by year-end, showing a trend toward predictable pricing. Outcome-based payment models are also gaining popularity, but face resistance from procurement departments preferring to compare fixed rates.

Why are agencies cutting staff while implementing AI

The marketing industry has cut approximately 8% of employees in 2025 and the first half of 2026, and S4 Capital has reduced headcount from 8,200 to 6,150 over four years. This is driven by two factors: automating part of operations and the need to control overall expenses amid rising technology costs. Forrester analyst warns that if real AI costs rise, automation may become more expensive than hiring people, putting agencies in a difficult position after already conducting layoffs.